Data

scRNA-seq data was captured from the bone marrow mononuclear cells of two patients with leukemia before and after undergoing a hematopoietic stem cell transplant. We subjected the cells from each patient to various dimensionality reduction techniques to determine whether they are capable of distinguishing the treatment states of the cells. scRNA-seq data from two healthy controls are used to perform cPCA and scPCA.

AML027

PCA

cPCA

scPCA

Analysis of Genes with Non-Zero Entries in Loadings Matrix

The following table provides the gene symbol and loadings of the genes with non-zero entries in one of the loadings vectors of the first two scPCs.

## # A tibble: 5 x 3
##   gene_sym   scPC1  scPC2
##   <chr>      <dbl>  <dbl>
## 1 CA1       0.0470  0    
## 2 STMN1    -0.997   0    
## 3 LDHA      0       0.482
## 4 C1QBP    -0.0606  0    
## 5 PDLIM1    0      -0.876

Comparison of Loadings: cPCA and scPCA

The absolute valules of the loadings are not compared for this patient, since only three entries in the leading loadings vectors produced by scPCA are non-zero.

cPCA (Tuned with Cross-Validation)

5-fold cross-validation was used to tune the contrastive parameter.

scPCA (Tuned with Cross-Validation)

5-fold cross-validation was used to tune the contrastive parameter.

t-SNE

With Initial PCA Step

Without Initial PCA Step

UMAP

ZINB-WaVE

SIMLR

Combined Plots

AML035

PCA

cPCA

scPCA

Analysis of Genes with Non-Zero Entries in Loadings Matrix

The following table provides the gene symbol and loadings of the genes with non-zero entries in one of the loadings vectors of the first two scPCs.

## # A tibble: 117 x 3
##    gene_sym    scPC1  scPC2
##    <chr>       <dbl>  <dbl>
##  1 HBB      -0.0618   0.712
##  2 HBA2      0       -0.117
##  3 HBA1      0.104   -0.660
##  4 RPL13    -0.0151   0    
##  5 RPL13A   -0.0126   0    
##  6 RPS4X    -0.0710   0    
##  7 PRDX2     0.0181   0    
##  8 PTMA      0.0787   0    
##  9 RPL19    -0.0687   0    
## 10 FTH1     -0.00350  0    
## # … with 107 more rows

Comparison of Loadings: cPCA and scPCA

cPCA (Tuned with Cross-Validation)

5-fold cross-validation was used to tune the contrastive parameter.

scPCA (Tuned with Cross-Validation)

5-fold cross-validation was used to tune the contrastive parameter.

t-SNE

With Initial PCA Step

Without Initial PCA Step

UMAP

ZINB-WaVE

SIMLR

Combined Plots

Running Time Analysis

The median running times over 5 repetitions of each method are presented below.